Robotic Lifestyle

ABB Group to sell ABB Robotics to SoftBank for $5.375B - The Robot Report
Humanoids·4 min read

Humanoids at the Edge: Why 2025’s Deals, Chips and Wireless Power Move Robots Out of the Lab

By Sophia Chen

From SoftBank’s $5.375 billion bid for ABB Robotics to new edge‑AI stacks and wireless charging that promises “100% uptime,” humanoid robots are moving from demonstrations toward everyday duty. The question now is engineering, not hype: can power, perception and safety be solved at scale so humanoids stop being prototypes and start being productive?

From SoftBank’s $5.375 billion bid for ABB Robotics to new edge‑AI stacks and wireless charging that promises “100% uptime,” humanoid robots are moving from demonstrations toward everyday duty. The question now is engineering, not hype: can power, perception and safety be solved at scale so humanoids stop being prototypes and start being productive?

A flurry of activity this fall—SoftBank’s acquisition offer for ABB Robotics (announced Oct. 8, 2025), NVIDIA‑powered platforms for multi‑domain robots, and CE approval for wireless power systems—has shifted the center of gravity in robotics toward integrated, production‑ready humanoid capabilities. Companies are combining industrial controls, high‑performance edge AI and new energy delivery to address the three hard constraints that held humanoids back: power, compute and safe interaction.

From lab stunt to factory shift: the business forces rearranging humanoids

That matters because the difference between a research demo and a deployed humanoid is measurable. ABB Robotics generated $2.3 billion in 2024 revenue and owns established manufacturing channels; SoftBank’s pitch—“physical AI,” Masayoshi Son said—signals money aiming to fuse industrial‑grade robotics with advanced AI. At the same time, nimble infrastructure plays—like CaPow’s Genesis wireless power (CE‑marked Oct. 2025), which claims fleets can be 30% smaller and achieve “100% uptime” for deployed AMRs—remove a key logistical barrier for mobile humanoids that need sustained operation in factories and warehouses.

From lab stunt to factory shift: the business forces rearranging humanoids

Humanoids have long been a magnet for headlines but a stubborn engineering problem. The new inflection comes from capital flows and strategic acquisitions. On Oct. 8, 2025, ABB announced it would sell its Robotics & Discrete Automation division to SoftBank for $5.375 billion—an explicit bet that robotics and AI together are the next industrial frontier. “SoftBank’s next frontier is physical AI,” Masayoshi Son declared, signaling appetite for deep R&D and commercialization.

The engineering triad: power, perception and platform

That matters because ABB brings manufacturing scale, field‑service networks and controllers such as OmniCore—assets that reduce go‑to‑market friction for robots built to industrial safety and uptime standards. For humanoid startups, an ABB+SoftBank axis could accelerate certification pathways, system integration and large‑batch deployments that small teams cannot finance alone. The likely timeline is practical: expect regulatory and antitrust reviews through 2026 and deal closure in mid‑to‑late 2026, after which commercialization roadmaps will crystallize.

Parallel to ownership moves, vendors are maturing the stack. YUAN’s announcement of NVIDIA Jetson Orin–based platforms and an ARC AI Platform aimed at humanoid‑style autonomy shows the compute side is being productized for real‑time, multi‑sensor fusion—LiDAR, IMU, GNSS, NVBLOX and Isaac ROS—so perception can run reliably at the edge where latency matters.

The engineering triad: power, perception and platform

Safety, TRL and the failure modes that still blink red

Humanoids require three simultaneous breakthroughs. First, power: conventional batteries constrain operating time and payload. CaPow’s Genesis system (CE‑marked in Oct. 2025) tackles this by delivering energy “in motion” to AMRs and claims fleets can shrink by roughly 30% because robots avoid charge downtime. CaPow CEO Mor M. Peretz said installations integrate in 10–20 minutes and “we guarantee 100% uptime to 100% of the fleet today,” underscoring how energy‑delivery logistics can alter robot economics. For humanoids that move and manipulate, continuous or opportunistic wireless charging removes a significant operational constraint.

Second, perception and compute: modern humanoids need sensor fusion and low‑latency inference. YUAN’s Jetson Orin–based platforms and NVIDIA’s Isaac ROS are designed for multi‑camera and LiDAR stacks and can run TensorRT‑accelerated networks at the edge. That reduces dependence on cloud connectivity and enables closed‑loop control for balance, grasping and dynamic walking—tasks that break if latency spikes.

Third, software and orchestration: fleet‑level intelligence (inspired by platforms like InOrbit and Energy Robotics) brings mission‑level LLM‑driven control, digital twins and evergreen mapping into play. Energy Robotics, for example, embeds LLM prompts into mission control to orchestrate inspections—parallels that humanoid fleets will borrow to translate high‑level commands into safe, low‑level motion plans.

Who wins, who loses—and the near-term battlegrounds

Safety, TRL and the failure modes that still blink red

Technical maturity varies by task. In structured factory work—repetitive assembly, fixed fixtures—humanoid capabilities are approaching TRL (Technology Readiness Level) 7–8 because environment variance is low and safety can be engineered through cages, force limits and deterministic controllers. But in unstructured human environments—homes, hospitals—critical capabilities like robust dexterous manipulation, long‑horizon planning and trustworthy natural‑language interaction sit closer to TRL 4–6.

Key failure modes remain concrete and remediable: power depletion mid‑task, slip‑and‑fall due to dynamic terrain, perception edge cases under rain or reflective surfaces, and LLM hallucinations driving unsafe behavior. Standards lag: ISO 10218 and ISO/TS 15066 govern industrial robots and collaborative cobots, while ISO 13482 covers personal care robots—yet a unified safety framework for semi‑autonomous humanoids operating near humans is still nascent. That gap raises certification and liability questions that buyers—manufacturers, logistics operators, care providers—will demand be answered before wide adoption.

Mitigations are practical: redundant power paths (onboard battery + opportunistic wireless charging), cross‑validated perception pipelines (LiDAR + stereo + IMU), motion envelopes with hard stops, and certifiable supervisory control that converts high‑level LLM intents into validated, bounded actions. Investors and integrators are already funding these layers because the ROI is simple: robots that stay powered, predictable and safe reduce downtime and labor risk—metrics CFOs understand.

Sources